Creativity Has Entered the Chat, With a Stranger: Novelty is a Nash Equilibrium

Kotaro Sakamoto, Shiro Takagi, Shuhei Ogawa, Yutaka Matsuo

Keywords: large language models, evolutionary game, multi-agent

Abstract: This study explores how novelty emerges in a research community by employing a multi-agent evolutionary game framework. The agents, representing scientists, engage in repeated interactions to generate and evaluate research ideas, following strategies of selfishness or cooperation. Our core hypothesis is that novelty is socially constructed, emerging through competition and dialogue. The game simulates a dynamic environment where agents balance diverging to promote their own ideas and merging to collaborate. Results suggest that agents with selfish strategies maximise the diversity of novel ideas, highlighting the tension between individual creativity and collective acceptance.

Submission Number: 34